基于时序循环图神经网络的电-气综合能源系统状态估计  

State estimation method for electric-gas integrated energy system based on time series cyclic graph neural network

作　　者：刘会家[1] 徐志高 徐智超 洪佳恩 张海伟 Liu Huijia;Xu Zhigao;Xu Zhichao;Hong Jia'en;Zhang Haiwei(School of Electrical and New Energy,China Three Gorges University,Yichang 443002,China;Jiangxi Provincial Port and Waterway Construction Investment Group,Nanchang 330008,China)

机构地区：[1]三峡大学电气与新能源学院,宜昌443002 [2]江西省港航建设投资集团有限公司,南昌330008

出　　处：《国外电子测量技术》2024年第12期34-42,共9页Foreign Electronic Measurement Technology

基　　金：国家自然科学基金(52277108)项目资助。

摘　　要：普通的神经网络模型在状态估计中容易受到系统拓扑结构变化和数据缺失的干扰,影响综合能源状态估计的精度。为了解决上述问题,提出一种将门控循环单元(gated recurrent unit,GRU)嵌入到能够深度提取邻域特征的图神经网络(graph neural network,GNN)模型,所提模型通过改进图神经网络的邻域矩阵使其成为自适应k阶邻域矩阵提高邻域特征信息提取能力,然后通过将GRU嵌入到图神经网络中考虑特征学习的时间和拓扑相关性的融合,使模型能够捕获更细粒度的网络拓扑的变化,从而提高模型抗干扰能力。算例仿真验证了模型与其他动态模型相比具有更好的估计精度和更强的鲁棒性。The ordinary neural network model is susceptible to the interference of system topology changes and data missing in state estimation,which affects the accuracy of integrated energy state estimation.In order to solve the above problems,a graph neural network(GNN)model that embeds a gated recurrent unit(GRU)into a graph neural network that can deeply extract neighborhood features is proposed.The proposed model improves the neighborhood matrix of the graph neural network to become an adaptive k-order neighborhood matrix to improve the ability of neighborhood feature information extraction.Then,by embedding GRU into the graph neural network to consider the fusion of time and topological correlation of feature learning,the model can capture more fine-grained network topology changes.So as to improve the anti-interference ability of the model.The simulation results show that the proposed model has better estimation accuracy and stronger robustness than other dynamic models.

关 键 词：状态估计 改进的图神经网络 k阶邻域 门控循环单元 综合能源系统 

分 类 号：TM743[电气工程—电力系统及自动化] TP312[自动化与计算机技术—计算机软件与理论] TN06[自动化与计算机技术—计算机科学与技术]